Social networks of lexical innovation

Load code

Variables

Load data

Filter data

success

Single lemma

Load data

Check tweets

tweets %>%
  select(tweet, date) %>%
  # slice(., sample(1:n())) #random selection
  arrange(date)

Check uses and edges

df_comp %>%
  filter(LEMMA == 'hyperlocal') %>%
  select(LEMMA, USES_TOT, SUBSET, USES)

Case studies

Data overview

Usage frequency

Overall across cases

df_comp %>%
  filter(
    LEMMA %in% cases,
    SUBSET == 'full'
    ) %>%
   arrange(desc(USES_TOT))

Uses in first subset

df_comp %>%
  filter(
    SUBSETTING == 'time',
    SUBSET == 'one'
    ) %>%
  select(LEMMA, USES_TOT, USES) %>%
  arrange(USES)

Degree centralization

Diachronic

plt <- df_comp %>%
  select(LEMMA, SUBSETTING, SUBSET, CENT_DEGREE, CENT_EV, DENSITY) %>%
  filter(
    SUBSET != 'full',
    LEMMA %in% cases
    # LEMMA %in% c(cases, 'hyperlocal', 'blockchain', 'climate denial', 'man bun', 'upskill', 'deep learning')
    # overtourism: diff too big for case study scale; 
    # , 'broflake', 'climate crisis', 'incel', 'overtourism'
  ) %>%
  ggplot(., aes(x=SUBSET, y=CENT_DEGREE)) + # group=1
    geom_point(aes(group=LEMMA, color=LEMMA, shape=LEMMA)) +
    geom_line(aes(group=LEMMA, color=LEMMA, linetype=LEMMA)) +
    guides(group=TRUE) +
    ggtitle('Diffusion over time: changes in degree centralization') +
    scale_y_continuous('degree centrality') +
    scale_x_discrete('subset')

ggplotly(plt)

# ggsave('out/cases_cent_diac.pdf', width=6, height=4)

Overall

df_comp %>%
  filter(
    LEMMA %in% cases,
    SUBSET == 'full'
    ) %>%
  select(
    LEMMA, 
    EDGES,
    CENT_DEGREE, 
    CENT_EV
    ) %>%
  arrange((CENT_DEGREE))

Full sample

Usage intensity

df_comp %>%
  filter(SUBSET == 'full') %>%
  arrange(desc(USES))

Edges

Degree centrality

Overall

List

df_comp %>%
  select(LEMMA, SUBSET, USES_TOT, CENT_DEGREE, CENT_EV) %>%
  filter(
    SUBSET == 'full'
    # USES >= 2
    ) %>%
  arrange(
    (CENT_DEGREE)
    # desc(CENT_EV)
  )

Plot

plt <- df_comp %>%
  select(LEMMA, SUBSET, USES, CENT_DEGREE) %>%
  filter(SUBSET == 'full') %>%
  arrange((CENT_DEGREE)) %>%
  ggplot(., aes(x=CENT_DEGREE, y=reorder(LEMMA, CENT_DEGREE))) +
    geom_point() +
    scale_y_discrete('lemmas') +
    scale_x_continuous(
      'degree centralization (log)',
      trans='log'
      )

plt

# ggsave('out/cent_sync_all.pdf', width=6, height=4)

Over time

Across all lemmas

df_comp %>%
  filter(
    SUBSET %in% c('one', 'two', 'three', 'four'),
    LEMMA %in% unsuccessful_diffs
    # USES_TOT >= 10000
    # USES > 1000
    ) %>%
  group_by(SUBSET) %>%
  summarize(
    DENS_AVG = mean(DENSITY),
    CENT_AVG = mean(CENT_DEGREE)
    ) %>%
  ggplot(., aes(x=SUBSET, y=DENS_AVG, group=1)) +
    geom_line() +
    geom_point() +
    scale_y_continuous('degree centralization') +
    scale_x_discrete('subsets')

# ggsave('out/full_cent_diac.pdf', width=6, height=4)

Biggest changes

df_comp %>%
  select(LEMMA, SUBSET, CENT_DEGREE, EDGES, USES_TOT) %>%
  filter(
    SUBSET %in% c(
      'one', 
      'four'
      ),
    USES_TOT >= 10000
    ) %>%
  dplyr::group_by(LEMMA) %>%
  dplyr::mutate(CENT_DIFF = CENT_DEGREE - lag(CENT_DEGREE, default=CENT_DEGREE[1])) %>%
  drop_na() %>%
  select(-SUBSET) %>%
  rename(
    CENT_LAST = CENT_DEGREE,
    EDGES_LAST = EDGES
    ) %>%
  arrange((CENT_DIFF))

Density

df_comp %>%
  select(LEMMA, SUBSET, USES_TOT, CENT_DEGREE, DENSITY) %>%
  filter(SUBSET == 'four') %>%
  arrange(DENSITY)

Frequency vs. networks

Frequency vs. centralization

Plot

plt <- df_comp %>%
  filter(
    SUBSET == 'four',
    # USES_TOT %in% (150000:500000)
    # LEMMA %in% c(cases)
    # !LEMMA %in% c('slut shaming', 'dashcam', 'shareable', 'cuckold', 'deep learning', 'hyperlocal')
    ) %>%
  select(LEMMA, CENT_DEGREE, USES_TOT, USES, EDGES) %>%
  ggplot(., aes(x=CENT_DEGREE, y=USES_TOT)) +
    geom_text(aes(label=LEMMA), hjust=-0.1, vjust=-0.1) + 
    # geom_point() +
    scale_y_continuous(
      'usage frequency (log)', 
      trans='log'
      ) +
    scale_x_continuous(
      'degree centralization'
      # trans='log'
      )

ggplotly(plt)

# ggsave('out/full_cent_freq_overall.pdf', width=6, height=4)
# ggsave('out/cases_cent_freq_overall.pdf', width=6, height=4)

Biggest discrepancies

df_comp %>%
  filter(
    SUBSETTING == 'time',
    SUBSET == 'four'
    ) %>%
  select(LEMMA, USES_TOT, CENT_DEGREE) %>%
  mutate(DISC = USES_TOT / CENT_DEGREE) %>%
  arrange(DISC)

Correlation

df_corr <- df_comp %>%
  filter(
    # SUBSET != 'full'
    # EDGES >= 100
    ) %>%
  select(-c(LEMMA, SUBSET, START, END, SKIP, STAMP))
  
cor.test(df_corr$USES, df_corr$CENT_DEGREE)

    Pearson's product-moment correlation

data:  df_corr$USES and df_corr$CENT_DEGREE
t = -2.3949, df = 517, p-value = 0.01698
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.18911153 -0.01884815
sample estimates:
       cor 
-0.1047473 

Coefficient of variation

df_comp %>%
  filter(
    SUBSET == 'full',
    USES_TOT >= 1000
    ) %>%
  select(LEMMA, USES_TOT, COEF_VAR) %>%
  arrange(desc(COEF_VAR))

Processing status

Lemma list

df_comp %>%
  select(LEMMA, SUBSET, STAMP) %>%
  filter(SUBSET == 'four') %>%
  # mutate(STAMP = as_datetime(STAMP)) %>%
  arrange(desc(STAMP))

Dataset statistics

df_comp %>%
  filter(SUBSET == 'full') %>%
  select(LEMMA, SUBSET, USES_TOT, USERS_TOT) %>%
  dplyr::summarise(
    USES_ALL = sum(USES_TOT),
    USERS_ALL = sum(USERS_TOT)
    )